Magnetic resonance (MR) imaging plays an increasingly
important role in the diagnosis and management of congenital heart
disease. Often, cardiovascular MR data are analyzed qualitatively.
Enhanced computing power and quantitative image analysis should
provide rapid, comprehensive and reproducible assessment of 4-dimensional
MR data sets.This proposal focuses on two groups of subjects -
postoperative tetralogy of Fallot patients and patients with connective
tissue disorders. These patients require accurate, serial assessment
of right ventricular function and aortic dimensions, respectively.
In this proposal, an image analysis methodology based on Active
Appearance Models (AAM) will be applied to both tasks. During
training, the AAM is built automatically from manually analyzed
image examples. In the analysis stage, the AAM allows fully automated
segmentation of image data using its learned knowledge of allowed
shapes and appearances of objects of interest - the ventricles
and the thoracic aorta.
Hypotheses and Aims:
Hypotheses driving this proposal are that
a) active appearance model-based segmentation can
provide automated, reproducible assessment of cardiovascular MR
images and increase the information content of these studies by
analyzing data in four dimensions (3-D + time), eliminating operator
variability and labor-intensive border tracing, and that
b) complete 4-D data sets of ventricular and aortic
surface morphology and motion will provide novel quantitative
indices of disease status.
We propose to:
